AI USE CASE
AI Sizing Recommendation Engine
Recommend the right size to online shoppers, reducing returns and boosting conversion.
What it is
Using machine learning on body measurements, purchase history, and return data, this engine predicts the optimal size for each customer across brands and styles. Retailers typically see return rates drop by 20–35% and conversion rates improve by 5–15% once the model is well-trained. By personalising size guidance at the product level, it also reduces customer frustration and repeat contacts to support. Over time, the model continuously refines predictions as new purchase and return signals accumulate.
Data you need
Historical purchase records, product return reasons, customer-provided body measurements or fit feedback, and SKU-level size charts across brands.
Required systems
- ecommerce platform
- crm
Why it works
- Collect structured return reasons at checkout or return portal to create a clean training signal.
- Standardise size chart ingestion across all catalogue brands before model training.
- Offer lightweight measurement capture (e.g. comparing to a garment that fits) to maximise data collection without friction.
- Retrain the model at least quarterly, aligned with new collection drops.
How this goes wrong
- Insufficient return reason data makes it impossible to distinguish size issues from other return causes, degrading model accuracy.
- Inconsistent brand size charts or missing product measurements cause poor cross-brand recommendations.
- Low customer uptake of measurement input (e.g. refusing to submit body data) limits personalisation.
- Model goes stale if not retrained after new seasonal collections or brand onboarding.
When NOT to do this
Don't implement a sizing engine if your catalogue has fewer than 500 SKUs or your return rate is already below 10% — the ROI won't justify the integration cost.
Vendors to consider
Sources
This use case is part of a larger Data & AI catalog built from 50+ enterprise transformation programs. Take the free diagnostic to see how it ranks against your specific context.